{"id":339405,"date":"2022-10-07T09:00:26","date_gmt":"2022-10-07T09:00:26","guid":{"rendered":"http:\/\/savepearlharbor.com\/?p=339405"},"modified":"-0001-11-30T00:00:00","modified_gmt":"-0001-11-29T21:00:00","slug":"","status":"publish","type":"post","link":"https:\/\/savepearlharbor.com\/?p=339405","title":{"rendered":"<span>\u0423\u0434\u0438\u0432\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u0440\u044f\u0434\u043e\u043c<\/span>"},"content":{"rendered":"<div><\/div>\n<div id=\"post-content-body\">\n<div>\n<div class=\"article-formatted-body article-formatted-body article-formatted-body_version-2\">\n<div xmlns=\"http:\/\/www.w3.org\/1999\/xhtml\">\n<p>\u0418\u043b\u0438 \u043d\u0435\u043e\u0431\u044b\u043a\u043d\u043e\u0432\u0435\u043d\u043d\u044b\u0435 \u0441\u0432\u043e\u0439\u0441\u0442\u0432\u0430 \u043e\u0431\u044b\u0447\u043d\u043e\u0439 U-net.<\/p>\n<p>\u041e\u0434\u043d\u043e\u0439 \u0438\u0437 \u0441\u0430\u043c\u044b\u0445 \u0436\u0443\u0442\u043a\u0438\u0445 \u043f\u0440\u043e\u0431\u043b\u0435\u043c \u0434\u043b\u044f \u043b\u044e\u0431\u043e\u0433\u043e \u043b\u044e\u0431\u0438\u0442\u0435\u043b\u044f, \u043a\u0430\u043a \u0438 \u0434\u043b\u044f \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u0430 \u0432 data science \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u043a\u0430\u0447\u0435\u0441\u0442\u0432\u043e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438. \u041a\u0430\u0447\u0435\u0441\u0442\u0432\u043e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438 \u0441\u043f\u043e\u0441\u043e\u0431\u043d\u043e \u043f\u043e\u0433\u0443\u0431\u0438\u0442\u044c \u0441\u0430\u043c\u0443\u044e \u0442\u043e\u043b\u043a\u043e\u0432\u0443\u044e \u0438 \u043a\u0440\u0430\u0441\u0438\u0432\u0443\u044e \u0438\u0434\u0435\u044e.<\/p>\n<p>\u041d\u043e \u043d\u0435 \u0432\u0441\u0451 \u043e\u043a\u0430\u0437\u0430\u043b\u043e\u0441\u044c \u0442\u0430\u043a \u043f\u043b\u043e\u0445\u043e \u0438 \u0432\u0430\u0448\u0435\u043c\u0443 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044e \u043f\u0440\u0435\u0434\u043b\u0430\u0433\u0430\u0435\u0442\u0441\u044f, \u043a\u0430\u043a \u0438 \u0432\u0441\u0435\u0433\u0434\u0430 \u0432 \u043c\u043e\u0438\u0445 \u043f\u043e\u0441\u0442\u0430\u0445, \u043a\u0440\u0430\u0441\u0438\u0432\u0430\u044f \u0438\u0434\u0435\u044f \u0441 \u043a\u043e\u0434\u0430\u043c\u0438 \u0438 \u043f\u0440\u0438\u043c\u0435\u0440\u043e\u043c.<\/p>\n<figure class=\"full-width\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/habrastorage.org\/r\/w1560\/getpro\/habr\/upload_files\/a4d\/bf1\/7fd\/a4dbf17fd7dc1f6ea619a8444ba470d4.png\" width=\"1080\" height=\"1080\" data-src=\"https:\/\/habrastorage.org\/getpro\/habr\/upload_files\/a4d\/bf1\/7fd\/a4dbf17fd7dc1f6ea619a8444ba470d4.png\"\/><figcaption><\/figcaption><\/figure>\n<hr\/>\n<p>\u0411\u0443\u0434\u0435\u043c \u0443\u0447\u0438\u0442\u044c \u0441\u0435\u0442\u044c \u043d\u0430\u0445\u043e\u0434\u0438\u0442\u044c \u043a\u0440\u0443\u0433 \u0432 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e\u0439 \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0435. <\/p>\n<p>\u0422.\u0435 \u0443 \u043d\u0430\u0441 \u0435\u0441\u0442\u044c \u043a\u0432\u0430\u0434\u0440\u0430\u0442, \u0437\u0430\u043f\u043e\u043b\u043d\u0435\u043d\u043d\u044b\u0439 \u0441\u043b\u0443\u0447\u0430\u0439\u043d\u044b\u043c\u0438 \u0442\u043e\u0447\u043a\u0430\u043c\u0438 \u0441 \u0437\u0430\u0440\u0430\u043d\u0435\u0435 \u0437\u0430\u0434\u0430\u043d\u043d\u044b\u043c\u0438 \u043f\u0430\u0440\u0430\u043c\u0435\u0442\u0440\u0430\u043c\u0438 \u0440\u0430\u0441\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u0438\u044f \u0438 \u0442\u0430\u043c \u0436\u0435 \u043a\u0440\u0443\u0433, \u043d\u043e \u0443\u0436\u0435 \u0441 \u0442\u043e\u0447\u043a\u0430\u043c\u0438 \u0438\u0437 \u0434\u0440\u0443\u0433\u043e\u0433\u043e \u0440\u0430\u0441\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u0438\u044f.  \u0421\u043e\u0437\u0434\u0430\u0434\u0438\u043c \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0442\u0430\u043a\u0436\u0435 \u043c\u0430\u0441\u043a\u0443 \u043e\u0431\u0443\u0447\u0430\u044e\u0449\u0443\u044e \u0438 \u043c\u0430\u0441\u043a\u0443 \u0438\u0441\u0442\u0438\u043d\u043d\u0443\u044e. <\/p>\n<p>\u0418 \u0442\u0430\u043a \u0443 \u043d\u0430\u0441 \u0435\u0441\u0442\u044c \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0430 \u0441 \u043a\u0440\u0443\u0433\u043e\u043c, \u043c\u0430\u0441\u043a\u0430 \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u044f\u0432\u043d\u043e \u043d\u0435 \u0441\u043e\u0432\u043f\u0430\u0434\u0430\u044e\u0449\u0430\u044f \u0441 \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u043e\u0439, \u0438 \u0442\u043e\u0447\u043d\u0430\u044f \u043c\u0430\u0441\u043a\u0430. \u0412\u043e\u0442 \u043f\u0440\u0438\u043c\u0435\u0440 \u043a\u0430\u0440\u0442\u0438\u043d\u043e\u043a. <\/p>\n<p>\u0414\u043b\u044f \u044d\u043a\u0441\u043f\u0435\u0440\u0438\u043c\u0435\u043d\u0442\u043e\u0432 \u0432\u043e\u0437\u044c\u043c\u0435\u043c \u0442\u0443 \u0436\u0435 \u0441\u0430\u043c\u0443\u044e, \u043e\u0447\u0435\u043d\u044c \u0445\u043e\u0440\u043e\u0448\u043e \u0438\u0437\u0443\u0447\u0435\u043d\u043d\u0443\u044e U-net. <\/p>\n<figure class=\"full-width\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/habrastorage.org\/r\/w1560\/getpro\/habr\/upload_files\/4d0\/91c\/c33\/4d091cc336ba60423b6c62bab9c8a0d7.png\" width=\"1551\" height=\"405\" data-src=\"https:\/\/habrastorage.org\/getpro\/habr\/upload_files\/4d0\/91c\/c33\/4d091cc336ba60423b6c62bab9c8a0d7.png\"\/><figcaption><\/figcaption><\/figure>\n<pre><code class=\"python\">import numpy as np import matplotlib.pyplot as plt %matplotlib inline import math from tqdm import tqdm_notebook, tqdm  import tensorflow as tf import keras as keras from keras import Model from keras.models import load_model from keras.optimizers import Adam from keras.layers import Input, Conv2D, Conv2DTranspose, MaxPooling2D, concatenate, Dropout from keras.losses import binary_crossentropy from keras.layers.core import Activation from keras import backend as K from keras.utils.generic_utils import get_custom_objects  from tqdm import tqdm_notebook  from math import sqrt<\/code><\/pre>\n<pre><code class=\"python\">w_size = 128 w2_size = w_size \/\/ 2 RR = int(w2_size * 0.5)  def next_pair(k):          delta = np.random.uniform(-10,10)     circle = np.zeros((w_size, w_size,1), dtype='int')     circle_mask = np.zeros((w_size, w_size), dtype='int')     R = RR - (np.random.random_sample()*10)     r_x = np.random.random_sample()*(RR\/\/2) # - R\/\/4     r_y = np.random.random_sample()*(RR\/\/2) # - R\/\/4     for i in range(w_size):         for j in range(w_size):             r = sqrt(float((i - w2_size - r_x)*(i - w2_size - r_x) +                            (j - w2_size - r_y)*(j - w2_size - r_y)))             # if r &lt; (R + np.random.uniform(-20,20)):             if r &lt; R + delta:                 circle_mask[i,j] = 1             if r &lt; R:                 circle[i,j,0] = 1     img_l = np.random.sample((w_size, w_size, 1))*0.5     img_h = np.random.sample((w_size, w_size, 1))*0.5 + 0.5      img = img_h.copy()     img[circle>0] = img_l[circle > 0]          msk = np.zeros((w_size, w_size, 1), dtype='float32')     msk[circle_mask>0] = 1. # \u043a\u0440\u0430\u0441\u0438\u043c \u043f\u0438\u043a\u0441\u0435\u043b\u0438 \u043c\u0430\u0441\u043a\u0438 \u044d\u043b\u043b\u0438\u043f\u0441\u0430      return img, msk, circle <\/code><\/pre>\n<p>\u0421\u043e\u0437\u0434\u0430\u0435\u043c \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0438, \u043c\u0430\u0441\u043a\u0438 \u0438 \u0438\u0441\u0442\u0438\u043d\u043d\u044b\u0435, \u0442\u043e\u0447\u043d\u044b\u0435 \u043c\u0430\u0441\u043a\u0438:<\/p>\n<pre><code class=\"python\">train_num = 2048 from joblib import Parallel, delayed train = np.array(Parallel(n_jobs=4)(delayed(next_pair)(k) for k in range(train_num))) train_x = train[:,0,:,:,:] train_y = train[:,1,:,:,:] train_r = train[:,2,:,:,:] # true mask <\/code><\/pre>\n<p>\u041d\u0435\u043c\u043d\u043e\u0433\u043e \u043c\u043e\u0434\u0438\u0444\u0438\u0446\u0438\u0440\u0443\u0435\u043c \u0441\u0435\u0442\u044c, DICE \u0442\u043e\u0447\u043d\u0435\u0435 \u0443\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u0441\u043e\u0432\u043f\u0430\u0434\u0435\u043d\u0438\u0435 \u043c\u0430\u0441\u043e\u043a:<\/p>\n<pre><code class=\"python\">def dice_coef(y_true, y_pred):     smooth = 1.     y_true_f = K.flatten(y_true)     y_pred_f = K.flatten(y_pred)     intersection = y_true_f * y_pred_f     score = (2. * K.sum(intersection) + smooth) \/ (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)     return score  get_custom_objects().update({'dice_coef': dice_coef })  def build_model(input_layer, start_neurons):     conv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(input_layer)     conv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(conv1)     pool1 = MaxPooling2D((2, 2))(conv1)     pool1 = Dropout(0.25)(pool1)      conv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(pool1)     conv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(conv2)     pool2 = MaxPooling2D((2, 2))(conv2)     pool2 = Dropout(0.5)(pool2)      conv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(pool2)     conv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(conv3)     pool3 = MaxPooling2D((2, 2))(conv3)     pool3 = Dropout(0.5)(pool3)      conv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(pool3)     conv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(conv4)     pool4 = MaxPooling2D((2, 2))(conv4)     pool4 = Dropout(0.5)(pool4)      # Middle     convm = Conv2D(start_neurons*16,(3,3),activation=\"relu\", padding=\"same\")(pool4)     convm = Conv2D(start_neurons*16,(3,3),activation=\"relu\", padding=\"same\")(convm)      deconv4 = Conv2DTranspose(start_neurons * 8, (3, 3), strides=(2, 2), padding=\"same\")(convm)     uconv4 = concatenate([deconv4, conv4])     uconv4 = Dropout(0.5)(uconv4)     uconv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(uconv4)     uconv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(uconv4)      deconv3 = Conv2DTranspose(start_neurons*4,(3,3),strides=(2, 2), padding=\"same\")(uconv4)     uconv3 = concatenate([deconv3, conv3])     uconv3 = Dropout(0.5)(uconv3)     uconv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(uconv3)     uconv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(uconv3)      deconv2 = Conv2DTranspose(start_neurons*2,(3,3),strides=(2, 2), padding=\"same\")(uconv3)     uconv2 = concatenate([deconv2, conv2])     uconv2 = Dropout(0.5)(uconv2)     uconv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(uconv2)     uconv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(uconv2)      deconv1 = Conv2DTranspose(start_neurons*1,(3,3),strides=(2, 2), padding=\"same\")(uconv2)     uconv1 = concatenate([deconv1, conv1])     uconv1 = Dropout(0.5)(uconv1)     uconv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(uconv1)     uconv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(uconv1)      uncov1 = Dropout(0.5)(uconv1)     # output_layer = Conv2D(1,(1,1), padding=\"same\", activation=\"sigmoid\")(uconv1)     output_layer = Conv2D(1,(1,1), padding=\"same\", activation=\"sigmoid\")(uconv1)          return output_layer  input_layer = Input((w_size, w_size, 1)) output_layer = build_model(input_layer, 16) model = Model(input_layer, output_layer) model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),               loss=tf.keras.losses.BinaryCrossentropy(),               metrics=['dice_coef'])  history = model.fit(train_x, train_y                     ,batch_size=16                     ,epochs=10                     ,verbose=2                     ,validation_split=0.2                     ,use_multiprocessing=True                    ) <\/code><\/pre>\n<p>\u0420\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442 \u043d\u0435 \u043e\u0447\u0435\u043d\u044c. \u0421\u0435\u0442\u044c \u043e\u0442\u043b\u0438\u0447\u043d\u043e \u0441\u043f\u0440\u0430\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0441 \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u0435\u043c \u043e\u0431\u043b\u0430\u0441\u0442\u0435\u0439 \u043f\u0440\u0438 \u0442\u043e\u0447\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0435.  \u041d\u043e \u0435\u0441\u043b\u0438 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0430 \u043f\u043b\u043e\u0445\u0430\u044f \u0438 \u0432\u0435\u043b\u0438\u043a\u0430 \u043e\u0448\u0438\u0431\u043a\u0430 \u043e\u0442 \u0440\u0435\u0430\u043b\u044c\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438, \u0442\u043e \u0438 \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u044c \u043d\u0430\u0448\u0435\u0439 \u0441\u0435\u0442\u0438 \u0431\u0443\u0434\u0435\u0442 \u043d\u0435 \u0438\u0434\u0435\u0430\u043b\u044c\u043d\u043e\u0439, \u0432\u0441\u0435\u0433\u043e 0.75. \u041a\u043e\u0433\u043e-\u0442\u043e \u043d\u0430\u0432\u0435\u0440\u043d\u043e \u044d\u0442\u043e \u0443\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u0435\u0442, \u043a\u043e\u043c\u0443-\u0442\u043e \u043c\u0430\u043b\u043e. <\/p>\n<pre><code class=\"python\">Epoch 1\/10 103\/103 - 3s - loss: 0.2875 - dice_coef: 0.4463 - val_loss: 0.1805 - val_dice_coef: 0.6473 Epoch 2\/10 103\/103 - 3s - loss: 0.1296 - dice_coef: 0.7414 - val_loss: 0.1174 - val_dice_coef: 0.7337 Epoch 3\/10 103\/103 - 3s - loss: 0.1162 - dice_coef: 0.7539 - val_loss: 0.1132 - val_dice_coef: 0.7482 Epoch 4\/10 103\/103 - 3s - loss: 0.1112 - dice_coef: 0.7603 - val_loss: 0.1091 - val_dice_coef: 0.7657 Epoch 5\/10 103\/103 - 3s - loss: 0.1085 - dice_coef: 0.7617 - val_loss: 0.1331 - val_dice_coef: 0.7584 Epoch 6\/10 103\/103 - 3s - loss: 0.1088 - dice_coef: 0.7605 - val_loss: 0.1080 - val_dice_coef: 0.7599 Epoch 7\/10 103\/103 - 3s - loss: 0.1064 - dice_coef: 0.7635 - val_loss: 0.1067 - val_dice_coef: 0.7573 Epoch 8\/10 103\/103 - 3s - loss: 0.1062 - dice_coef: 0.7638 - val_loss: 0.1069 - val_dice_coef: 0.7587 Epoch 9\/10 103\/103 - 3s - loss: 0.1054 - dice_coef: 0.7649 - val_loss: 0.1068 - val_dice_coef: 0.7617 Epoch 10\/10 103\/103 - 3s - loss: 0.1058 - dice_coef: 0.7644 - val_loss: 0.1087 - val_dice_coef: 0.7616 1 <\/code><\/pre>\n<p>\u041d\u043e \u043d\u0430\u0441 \u0438\u043d\u0442\u0435\u0440\u0435\u0441\u0443\u0435\u0442 \u0434\u0440\u0443\u0433\u043e\u0435, \u0433\u043e\u0440\u0430\u0437\u0434\u043e \u0431\u043e\u043b\u0435\u0435 \u0438\u043d\u0442\u0435\u0440\u0435\u0441\u043d\u043e\u0435 \u044f\u0432\u043b\u0435\u043d\u0438\u0435. \u0421\u0440\u0430\u0432\u043d\u0438\u043c \u043d\u0430\u0448\u0443 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0438 \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435. <\/p>\n<pre><code>pred = model.predict(train_x) dice_coef(train_y.astype('float32'), pred).numpy()   0.7675821 dice_coef(train_r.astype('float32'), pred).numpy()   0.819669<\/code><\/pre>\n<p>\u0418 \u0442\u0443\u0442 \u0432\u0434\u0440\u0443\u0433 \u0432\u043e\u0442 \u043e\u043d\u043e &#8212; \u043e\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442\u0441\u044f \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435 \u043c\u0435\u043d\u044c\u0448\u0435 \u043e\u0442\u043b\u0438\u0447\u0430\u0435\u0442\u0441\u044f \u043e\u0442 \u0438\u0441\u0442\u0438\u043d\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438, \u043d\u0435\u0436\u0435\u043b\u0438 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u043d\u0430\u044f \u043d\u0430\u043c\u0438 \u0438\u0441\u043a\u0430\u0436\u0435\u043d\u043d\u0430\u044f \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0430. \u041e\u0442\u043b\u0438\u0447\u043d\u043e. <\/p>\n<p>\u041c\u044b \u0442\u0435\u043f\u0435\u0440\u044c \u0432\u044b\u0431\u0440\u0430\u0441\u044b\u0432\u0430\u0435\u043c \u043d\u0430\u0448\u0443 \u043f\u0435\u0440\u0432\u043e\u043d\u0430\u0447\u0430\u043b\u044c\u043d\u0443\u044e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0437\u0430 \u043d\u0435\u043d\u0430\u0434\u043e\u0431\u043d\u043e\u0441\u0442\u044c\u044e \u0438 \u043f\u0440\u043e\u0432\u043e\u0434\u0438\u043c \u043d\u043e\u0432\u044b\u0439 \u0441\u0435\u0430\u043d\u0441 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u0443\u0436\u0435 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u044f \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435 pred \u043a\u0430\u043a \u043d\u043e\u0432\u0443\u044e \u043c\u0430\u0441\u043a\u0443.  <\/p>\n<pre><code class=\"python\">model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),               loss=tf.keras.losses.BinaryCrossentropy(),               metrics=['dice_coef']) history = model.fit(train_x, pred>0.5                     ,batch_size=16                     ,epochs=10                     ,verbose=2                     ,validation_split=0.2                     ,use_multiprocessing=True                    ) <\/code><\/pre>\n<pre><code class=\"python\">Epoch 1\/10 103\/103 - 3s - loss: 0.0045 - dice_coef: 0.9899 - val_loss: 0.0021 - val_dice_coef: 0.9944 Epoch 2\/10 103\/103 - 3s - loss: 0.0030 - dice_coef: 0.9933 - val_loss: 0.0017 - val_dice_coef: 0.9953 Epoch 3\/10 103\/103 - 3s - loss: 0.0027 - dice_coef: 0.9941 - val_loss: 0.0016 - val_dice_coef: 0.9958 Epoch 4\/10 103\/103 - 3s - loss: 0.0024 - dice_coef: 0.9946 - val_loss: 0.0016 - val_dice_coef: 0.9960 Epoch 5\/10 103\/103 - 3s - loss: 0.0022 - dice_coef: 0.9951 - val_loss: 0.0014 - val_dice_coef: 0.9963 Epoch 6\/10 103\/103 - 3s - loss: 0.0021 - dice_coef: 0.9953 - val_loss: 0.0014 - val_dice_coef: 0.9963 Epoch 7\/10 103\/103 - 3s - loss: 0.0021 - dice_coef: 0.9955 - val_loss: 0.0014 - val_dice_coef: 0.9965 Epoch 8\/10 103\/103 - 3s - loss: 0.0020 - dice_coef: 0.9957 - val_loss: 0.0014 - val_dice_coef: 0.9965 Epoch 9\/10 103\/103 - 3s - loss: 0.0019 - dice_coef: 0.9958 - val_loss: 0.0013 - val_dice_coef: 0.9967 Epoch 10\/10 103\/103 - 3s - loss: 0.0019 - dice_coef: 0.9959 - val_loss: 0.0014 - val_dice_coef: 0.9965<\/code><\/pre>\n<p>\u0418 \u0442\u0443\u0442 \u043f\u043e\u043b\u0443\u0447\u0430\u0435\u043c \u043f\u0440\u043e\u0441\u0442\u043e \u0444\u0430\u043d\u0442\u0430\u0441\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0439 \u0440\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442 &#8212; \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u044c \u0431\u043e\u043b\u044c\u0448\u0435 0.99! <\/p>\n<p>\u0410 \u043a\u043e\u043d\u043a\u0440\u0435\u0442\u043d\u043e \u0434\u043e\u0441\u0442\u0438\u0433\u0430\u0435\u043c dice_coeff 0.9919946, \u0438\u043c\u0435\u044f \u043d\u0430 \u0440\u0443\u043a\u0430\u0445 \u0434\u0440\u044f\u043d\u043d\u0443\u044e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0432 \u043d\u0430\u0447\u0430\u043b\u0435.<\/p>\n<pre><code>pred_1 = model.predict(train_x) dice_coef(train_r.astype('float32'), pred_1).numpy()   0.9919946 <\/code><\/pre>\n<p>\u0418\u0442\u0430\u043a \u043c\u044b \u043f\u043e\u043b\u0443\u0447\u0438\u043b\u0438 \u0441\u043b\u0435\u0434\u0443\u044e\u0449\u0435\u0435:  \u0443 \u043d\u0430\u0441 \u0435\u0441\u0442\u044c \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0438 \u0434\u043b\u044f \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0438 \u043d\u0435\u043a\u0430\u0447\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u0430\u044f \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0430. <\/p>\n<p>\u041f\u0440\u043e\u044f\u0432\u0438\u0432 \u0440\u0435\u0430\u043b\u044c\u043d\u044b\u0439 \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442, \u0441\u043c\u0435\u043a\u0430\u043b\u043a\u0443 \u0438 \u043d\u0435\u043c\u043d\u043e\u0433\u043e \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0433\u043e \u0438\u043d\u0442\u0435\u043b\u043b\u0435\u043a\u0442\u0430, \u043c\u044b \u0432\u043e\u0441\u0441\u0442\u0430\u043d\u043e\u0432\u0438\u043b\u0438 \u0440\u0435\u0430\u043b\u044c\u043d\u0443\u044e, \u0442\u043e\u0447\u043d\u0443\u044e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443, \u0438\u0437 \u0434\u0430\u043d\u043d\u043e\u0439 \u043d\u0430\u043c \u043d\u0435\u043a\u0430\u0447\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0439. \u0422.\u0435. \u043d\u0430\u0448\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u0430\u0446\u0438\u0438 \u0432\u0441\u0451-\u0442\u0430\u043a\u0438 \u0434\u043e\u0441\u0442\u0438\u0433\u043b\u043e \u043d\u0443\u0436\u043d\u043e\u0439 \u0438 \u043f\u0440\u0438\u0435\u043c\u043b\u0435\u043c\u043e\u0439 \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u0438. <\/p>\n<p>\u041a\u043e\u043d\u0435\u0447\u043d\u043e \u0436\u0435, \u0443 \u043d\u0430\u0441 \u0438\u0433\u0440\u0443\u0448\u0435\u0447\u043d\u044b\u0439 \u0434\u0430\u0442\u0430\u0441\u0435\u0442, \u043f\u0440\u043e\u0441\u0442\u0430\u044f U-net \u0438 \u043c\u044b \u043d\u0435 \u043f\u044b\u0442\u0430\u0435\u043c\u0441\u044f \u0440\u0435\u0448\u0430\u0442\u044c \u0433\u043b\u043e\u0431\u0430\u043b\u044c\u043d\u044b\u0435 \u0437\u0430\u0434\u0430\u0447\u0438 \u0438 \u0431\u044b\u0442\u044c \u0442\u0430\u043c, \u0433\u0434\u0435 \u043d\u0435 \u0445\u043e\u0442\u0438\u043c \u0431\u044b\u0442\u044c.  <\/p>\n<p>\u041d\u043e \u0438\u0434\u0435\u044f \u043f\u043e\u043b\u0435\u0437\u043d\u0430\u044f, \u043d\u0443\u0436\u043d\u0430\u044f \u0438 \u043f\u0440\u0438\u0433\u043e\u0434\u0438\u0442\u0441\u044f \u043c\u043e\u0436\u0435\u0442 \u0432\u0441\u0435\u043c.<\/p>\n<p>\u0413\u043b\u0430\u0432\u043d\u043e\u0435, \u044d\u0442\u043e \u043f\u043e\u0432\u0435\u0440\u0438\u0442\u044c, \u0447\u0442\u043e \u0438 \u043d\u0430 \u0432\u0430\u0448\u0435\u043c \u0434\u0430\u0442\u0430\u0441\u0435\u0442\u0435, \u0441 \u0432\u0430\u0448\u0435\u0439 \u0441\u0435\u0442\u044c\u044e \u044d\u0442\u043e\u0442 \u044d\u0444\u0444\u0435\u043a\u0442 \u0441\u043e\u0445\u0440\u0430\u043d\u0438\u0442\u0441\u044f )) <\/p>\n<\/p>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"v-portal\" style=\"display:none;\"><\/div>\n<\/div>\n<p> <!----> <!----><br \/> \u0441\u0441\u044b\u043b\u043a\u0430 \u043d\u0430 \u043e\u0440\u0438\u0433\u0438\u043d\u0430\u043b \u0441\u0442\u0430\u0442\u044c\u0438 <a href=\"https:\/\/habr.com\/ru\/post\/691952\/\"> https:\/\/habr.com\/ru\/post\/691952\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<div><\/div>\n<div id=\"post-content-body\">\n<div>\n<div class=\"article-formatted-body article-formatted-body article-formatted-body_version-2\">\n<div xmlns=\"http:\/\/www.w3.org\/1999\/xhtml\">\n<p>\u0418\u043b\u0438 \u043d\u0435\u043e\u0431\u044b\u043a\u043d\u043e\u0432\u0435\u043d\u043d\u044b\u0435 \u0441\u0432\u043e\u0439\u0441\u0442\u0432\u0430 \u043e\u0431\u044b\u0447\u043d\u043e\u0439 U-net.<\/p>\n<p>\u041e\u0434\u043d\u043e\u0439 \u0438\u0437 \u0441\u0430\u043c\u044b\u0445 \u0436\u0443\u0442\u043a\u0438\u0445 \u043f\u0440\u043e\u0431\u043b\u0435\u043c \u0434\u043b\u044f \u043b\u044e\u0431\u043e\u0433\u043e \u043b\u044e\u0431\u0438\u0442\u0435\u043b\u044f, \u043a\u0430\u043a \u0438 \u0434\u043b\u044f \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u0430 \u0432 data science \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u043a\u0430\u0447\u0435\u0441\u0442\u0432\u043e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438. \u041a\u0430\u0447\u0435\u0441\u0442\u0432\u043e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438 \u0441\u043f\u043e\u0441\u043e\u0431\u043d\u043e \u043f\u043e\u0433\u0443\u0431\u0438\u0442\u044c \u0441\u0430\u043c\u0443\u044e \u0442\u043e\u043b\u043a\u043e\u0432\u0443\u044e \u0438 \u043a\u0440\u0430\u0441\u0438\u0432\u0443\u044e \u0438\u0434\u0435\u044e.<\/p>\n<p>\u041d\u043e \u043d\u0435 \u0432\u0441\u0451 \u043e\u043a\u0430\u0437\u0430\u043b\u043e\u0441\u044c \u0442\u0430\u043a \u043f\u043b\u043e\u0445\u043e \u0438 \u0432\u0430\u0448\u0435\u043c\u0443 \u0432\u043d\u0438\u043c\u0430\u043d\u0438\u044e \u043f\u0440\u0435\u0434\u043b\u0430\u0433\u0430\u0435\u0442\u0441\u044f, \u043a\u0430\u043a \u0438 \u0432\u0441\u0435\u0433\u0434\u0430 \u0432 \u043c\u043e\u0438\u0445 \u043f\u043e\u0441\u0442\u0430\u0445, \u043a\u0440\u0430\u0441\u0438\u0432\u0430\u044f \u0438\u0434\u0435\u044f \u0441 \u043a\u043e\u0434\u0430\u043c\u0438 \u0438 \u043f\u0440\u0438\u043c\u0435\u0440\u043e\u043c.<\/p>\n<figure class=\"full-width\"><figcaption><\/figcaption><\/figure>\n<hr\/>\n<p>\u0411\u0443\u0434\u0435\u043c \u0443\u0447\u0438\u0442\u044c \u0441\u0435\u0442\u044c \u043d\u0430\u0445\u043e\u0434\u0438\u0442\u044c \u043a\u0440\u0443\u0433 \u0432 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e\u0439 \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0435. <\/p>\n<p>\u0422.\u0435 \u0443 \u043d\u0430\u0441 \u0435\u0441\u0442\u044c \u043a\u0432\u0430\u0434\u0440\u0430\u0442, \u0437\u0430\u043f\u043e\u043b\u043d\u0435\u043d\u043d\u044b\u0439 \u0441\u043b\u0443\u0447\u0430\u0439\u043d\u044b\u043c\u0438 \u0442\u043e\u0447\u043a\u0430\u043c\u0438 \u0441 \u0437\u0430\u0440\u0430\u043d\u0435\u0435 \u0437\u0430\u0434\u0430\u043d\u043d\u044b\u043c\u0438 \u043f\u0430\u0440\u0430\u043c\u0435\u0442\u0440\u0430\u043c\u0438 \u0440\u0430\u0441\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u0438\u044f \u0438 \u0442\u0430\u043c \u0436\u0435 \u043a\u0440\u0443\u0433, \u043d\u043e \u0443\u0436\u0435 \u0441 \u0442\u043e\u0447\u043a\u0430\u043c\u0438 \u0438\u0437 \u0434\u0440\u0443\u0433\u043e\u0433\u043e \u0440\u0430\u0441\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u0438\u044f.  \u0421\u043e\u0437\u0434\u0430\u0434\u0438\u043c \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u0442\u0430\u043a\u0436\u0435 \u043c\u0430\u0441\u043a\u0443 \u043e\u0431\u0443\u0447\u0430\u044e\u0449\u0443\u044e \u0438 \u043c\u0430\u0441\u043a\u0443 \u0438\u0441\u0442\u0438\u043d\u043d\u0443\u044e. <\/p>\n<p>\u0418 \u0442\u0430\u043a \u0443 \u043d\u0430\u0441 \u0435\u0441\u0442\u044c \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0430 \u0441 \u043a\u0440\u0443\u0433\u043e\u043c, \u043c\u0430\u0441\u043a\u0430 \u0434\u043b\u044f \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u044f\u0432\u043d\u043e \u043d\u0435 \u0441\u043e\u0432\u043f\u0430\u0434\u0430\u044e\u0449\u0430\u044f \u0441 \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u043e\u0439, \u0438 \u0442\u043e\u0447\u043d\u0430\u044f \u043c\u0430\u0441\u043a\u0430. \u0412\u043e\u0442 \u043f\u0440\u0438\u043c\u0435\u0440 \u043a\u0430\u0440\u0442\u0438\u043d\u043e\u043a. <\/p>\n<p>\u0414\u043b\u044f \u044d\u043a\u0441\u043f\u0435\u0440\u0438\u043c\u0435\u043d\u0442\u043e\u0432 \u0432\u043e\u0437\u044c\u043c\u0435\u043c \u0442\u0443 \u0436\u0435 \u0441\u0430\u043c\u0443\u044e, \u043e\u0447\u0435\u043d\u044c \u0445\u043e\u0440\u043e\u0448\u043e \u0438\u0437\u0443\u0447\u0435\u043d\u043d\u0443\u044e U-net. <\/p>\n<figure class=\"full-width\"><figcaption><\/figcaption><\/figure>\n<pre><code class=\"python\">import numpy as np import matplotlib.pyplot as plt %matplotlib inline import math from tqdm import tqdm_notebook, tqdm  import tensorflow as tf import keras as keras from keras import Model from keras.models import load_model from keras.optimizers import Adam from keras.layers import Input, Conv2D, Conv2DTranspose, MaxPooling2D, concatenate, Dropout from keras.losses import binary_crossentropy from keras.layers.core import Activation from keras import backend as K from keras.utils.generic_utils import get_custom_objects  from tqdm import tqdm_notebook  from math import sqrt<\/code><\/pre>\n<pre><code class=\"python\">w_size = 128 w2_size = w_size \/\/ 2 RR = int(w2_size * 0.5)  def next_pair(k):          delta = np.random.uniform(-10,10)     circle = np.zeros((w_size, w_size,1), dtype='int')     circle_mask = np.zeros((w_size, w_size), dtype='int')     R = RR - (np.random.random_sample()*10)     r_x = np.random.random_sample()*(RR\/\/2) # - R\/\/4     r_y = np.random.random_sample()*(RR\/\/2) # - R\/\/4     for i in range(w_size):         for j in range(w_size):             r = sqrt(float((i - w2_size - r_x)*(i - w2_size - r_x) +                            (j - w2_size - r_y)*(j - w2_size - r_y)))             # if r &lt; (R + np.random.uniform(-20,20)):             if r &lt; R + delta:                 circle_mask[i,j] = 1             if r &lt; R:                 circle[i,j,0] = 1     img_l = np.random.sample((w_size, w_size, 1))*0.5     img_h = np.random.sample((w_size, w_size, 1))*0.5 + 0.5      img = img_h.copy()     img[circle>0] = img_l[circle > 0]          msk = np.zeros((w_size, w_size, 1), dtype='float32')     msk[circle_mask>0] = 1. # \u043a\u0440\u0430\u0441\u0438\u043c \u043f\u0438\u043a\u0441\u0435\u043b\u0438 \u043c\u0430\u0441\u043a\u0438 \u044d\u043b\u043b\u0438\u043f\u0441\u0430      return img, msk, circle <\/code><\/pre>\n<p>\u0421\u043e\u0437\u0434\u0430\u0435\u043c \u043a\u0430\u0440\u0442\u0438\u043d\u043a\u0438, \u043c\u0430\u0441\u043a\u0438 \u0438 \u0438\u0441\u0442\u0438\u043d\u043d\u044b\u0435, \u0442\u043e\u0447\u043d\u044b\u0435 \u043c\u0430\u0441\u043a\u0438:<\/p>\n<pre><code class=\"python\">train_num = 2048 from joblib import Parallel, delayed train = np.array(Parallel(n_jobs=4)(delayed(next_pair)(k) for k in range(train_num))) train_x = train[:,0,:,:,:] train_y = train[:,1,:,:,:] train_r = train[:,2,:,:,:] # true mask <\/code><\/pre>\n<p>\u041d\u0435\u043c\u043d\u043e\u0433\u043e \u043c\u043e\u0434\u0438\u0444\u0438\u0446\u0438\u0440\u0443\u0435\u043c \u0441\u0435\u0442\u044c, DICE \u0442\u043e\u0447\u043d\u0435\u0435 \u0443\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442 \u0441\u043e\u0432\u043f\u0430\u0434\u0435\u043d\u0438\u0435 \u043c\u0430\u0441\u043e\u043a:<\/p>\n<pre><code class=\"python\">def dice_coef(y_true, y_pred):     smooth = 1.     y_true_f = K.flatten(y_true)     y_pred_f = K.flatten(y_pred)     intersection = y_true_f * y_pred_f     score = (2. * K.sum(intersection) + smooth) \/ (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)     return score  get_custom_objects().update({'dice_coef': dice_coef })  def build_model(input_layer, start_neurons):     conv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(input_layer)     conv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(conv1)     pool1 = MaxPooling2D((2, 2))(conv1)     pool1 = Dropout(0.25)(pool1)      conv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(pool1)     conv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(conv2)     pool2 = MaxPooling2D((2, 2))(conv2)     pool2 = Dropout(0.5)(pool2)      conv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(pool2)     conv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(conv3)     pool3 = MaxPooling2D((2, 2))(conv3)     pool3 = Dropout(0.5)(pool3)      conv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(pool3)     conv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(conv4)     pool4 = MaxPooling2D((2, 2))(conv4)     pool4 = Dropout(0.5)(pool4)      # Middle     convm = Conv2D(start_neurons*16,(3,3),activation=\"relu\", padding=\"same\")(pool4)     convm = Conv2D(start_neurons*16,(3,3),activation=\"relu\", padding=\"same\")(convm)      deconv4 = Conv2DTranspose(start_neurons * 8, (3, 3), strides=(2, 2), padding=\"same\")(convm)     uconv4 = concatenate([deconv4, conv4])     uconv4 = Dropout(0.5)(uconv4)     uconv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(uconv4)     uconv4 = Conv2D(start_neurons*8,(3,3),activation=\"relu\", padding=\"same\")(uconv4)      deconv3 = Conv2DTranspose(start_neurons*4,(3,3),strides=(2, 2), padding=\"same\")(uconv4)     uconv3 = concatenate([deconv3, conv3])     uconv3 = Dropout(0.5)(uconv3)     uconv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(uconv3)     uconv3 = Conv2D(start_neurons*4,(3,3),activation=\"relu\", padding=\"same\")(uconv3)      deconv2 = Conv2DTranspose(start_neurons*2,(3,3),strides=(2, 2), padding=\"same\")(uconv3)     uconv2 = concatenate([deconv2, conv2])     uconv2 = Dropout(0.5)(uconv2)     uconv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(uconv2)     uconv2 = Conv2D(start_neurons*2,(3,3),activation=\"relu\", padding=\"same\")(uconv2)      deconv1 = Conv2DTranspose(start_neurons*1,(3,3),strides=(2, 2), padding=\"same\")(uconv2)     uconv1 = concatenate([deconv1, conv1])     uconv1 = Dropout(0.5)(uconv1)     uconv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(uconv1)     uconv1 = Conv2D(start_neurons*1,(3,3),activation=\"relu\", padding=\"same\")(uconv1)      uncov1 = Dropout(0.5)(uconv1)     # output_layer = Conv2D(1,(1,1), padding=\"same\", activation=\"sigmoid\")(uconv1)     output_layer = Conv2D(1,(1,1), padding=\"same\", activation=\"sigmoid\")(uconv1)          return output_layer  input_layer = Input((w_size, w_size, 1)) output_layer = build_model(input_layer, 16) model = Model(input_layer, output_layer) model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),               loss=tf.keras.losses.BinaryCrossentropy(),               metrics=['dice_coef'])  history = model.fit(train_x, train_y                     ,batch_size=16                     ,epochs=10                     ,verbose=2                     ,validation_split=0.2                     ,use_multiprocessing=True                    ) <\/code><\/pre>\n<p>\u0420\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442 \u043d\u0435 \u043e\u0447\u0435\u043d\u044c. \u0421\u0435\u0442\u044c \u043e\u0442\u043b\u0438\u0447\u043d\u043e \u0441\u043f\u0440\u0430\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0441 \u0440\u0430\u0441\u043f\u043e\u0437\u043d\u0430\u0432\u0430\u043d\u0438\u0435\u043c \u043e\u0431\u043b\u0430\u0441\u0442\u0435\u0439 \u043f\u0440\u0438 \u0442\u043e\u0447\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0435.  \u041d\u043e \u0435\u0441\u043b\u0438 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0430 \u043f\u043b\u043e\u0445\u0430\u044f \u0438 \u0432\u0435\u043b\u0438\u043a\u0430 \u043e\u0448\u0438\u0431\u043a\u0430 \u043e\u0442 \u0440\u0435\u0430\u043b\u044c\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438, \u0442\u043e \u0438 \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u044c \u043d\u0430\u0448\u0435\u0439 \u0441\u0435\u0442\u0438 \u0431\u0443\u0434\u0435\u0442 \u043d\u0435 \u0438\u0434\u0435\u0430\u043b\u044c\u043d\u043e\u0439, \u0432\u0441\u0435\u0433\u043e 0.75. \u041a\u043e\u0433\u043e-\u0442\u043e \u043d\u0430\u0432\u0435\u0440\u043d\u043e \u044d\u0442\u043e \u0443\u0441\u0442\u0440\u0430\u0438\u0432\u0430\u0435\u0442, \u043a\u043e\u043c\u0443-\u0442\u043e \u043c\u0430\u043b\u043e. <\/p>\n<pre><code class=\"python\">Epoch 1\/10 103\/103 - 3s - loss: 0.2875 - dice_coef: 0.4463 - val_loss: 0.1805 - val_dice_coef: 0.6473 Epoch 2\/10 103\/103 - 3s - loss: 0.1296 - dice_coef: 0.7414 - val_loss: 0.1174 - val_dice_coef: 0.7337 Epoch 3\/10 103\/103 - 3s - loss: 0.1162 - dice_coef: 0.7539 - val_loss: 0.1132 - val_dice_coef: 0.7482 Epoch 4\/10 103\/103 - 3s - loss: 0.1112 - dice_coef: 0.7603 - val_loss: 0.1091 - val_dice_coef: 0.7657 Epoch 5\/10 103\/103 - 3s - loss: 0.1085 - dice_coef: 0.7617 - val_loss: 0.1331 - val_dice_coef: 0.7584 Epoch 6\/10 103\/103 - 3s - loss: 0.1088 - dice_coef: 0.7605 - val_loss: 0.1080 - val_dice_coef: 0.7599 Epoch 7\/10 103\/103 - 3s - loss: 0.1064 - dice_coef: 0.7635 - val_loss: 0.1067 - val_dice_coef: 0.7573 Epoch 8\/10 103\/103 - 3s - loss: 0.1062 - dice_coef: 0.7638 - val_loss: 0.1069 - val_dice_coef: 0.7587 Epoch 9\/10 103\/103 - 3s - loss: 0.1054 - dice_coef: 0.7649 - val_loss: 0.1068 - val_dice_coef: 0.7617 Epoch 10\/10 103\/103 - 3s - loss: 0.1058 - dice_coef: 0.7644 - val_loss: 0.1087 - val_dice_coef: 0.7616 1 <\/code><\/pre>\n<p>\u041d\u043e \u043d\u0430\u0441 \u0438\u043d\u0442\u0435\u0440\u0435\u0441\u0443\u0435\u0442 \u0434\u0440\u0443\u0433\u043e\u0435, \u0433\u043e\u0440\u0430\u0437\u0434\u043e \u0431\u043e\u043b\u0435\u0435 \u0438\u043d\u0442\u0435\u0440\u0435\u0441\u043d\u043e\u0435 \u044f\u0432\u043b\u0435\u043d\u0438\u0435. \u0421\u0440\u0430\u0432\u043d\u0438\u043c \u043d\u0430\u0448\u0443 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0438 \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435. <\/p>\n<pre><code>pred = model.predict(train_x) dice_coef(train_y.astype('float32'), pred).numpy()   0.7675821 dice_coef(train_r.astype('float32'), pred).numpy()   0.819669<\/code><\/pre>\n<p>\u0418 \u0442\u0443\u0442 \u0432\u0434\u0440\u0443\u0433 \u0432\u043e\u0442 \u043e\u043d\u043e &#8212; \u043e\u043a\u0430\u0437\u044b\u0432\u0430\u0435\u0442\u0441\u044f \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435 \u043c\u0435\u043d\u044c\u0448\u0435 \u043e\u0442\u043b\u0438\u0447\u0430\u0435\u0442\u0441\u044f \u043e\u0442 \u0438\u0441\u0442\u0438\u043d\u043d\u043e\u0439 \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0438, \u043d\u0435\u0436\u0435\u043b\u0438 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u043d\u0430\u044f \u043d\u0430\u043c\u0438 \u0438\u0441\u043a\u0430\u0436\u0435\u043d\u043d\u0430\u044f \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0430. \u041e\u0442\u043b\u0438\u0447\u043d\u043e. <\/p>\n<p>\u041c\u044b \u0442\u0435\u043f\u0435\u0440\u044c \u0432\u044b\u0431\u0440\u0430\u0441\u044b\u0432\u0430\u0435\u043c \u043d\u0430\u0448\u0443 \u043f\u0435\u0440\u0432\u043e\u043d\u0430\u0447\u0430\u043b\u044c\u043d\u0443\u044e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0437\u0430 \u043d\u0435\u043d\u0430\u0434\u043e\u0431\u043d\u043e\u0441\u0442\u044c\u044e \u0438 \u043f\u0440\u043e\u0432\u043e\u0434\u0438\u043c \u043d\u043e\u0432\u044b\u0439 \u0441\u0435\u0430\u043d\u0441 \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f, \u0443\u0436\u0435 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u0443\u044f \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u043d\u043e\u0435 \u043f\u0440\u0435\u0434\u0441\u043a\u0430\u0437\u0430\u043d\u0438\u0435 pred \u043a\u0430\u043a \u043d\u043e\u0432\u0443\u044e \u043c\u0430\u0441\u043a\u0443.  <\/p>\n<pre><code class=\"python\">model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),               loss=tf.keras.losses.BinaryCrossentropy(),               metrics=['dice_coef']) history = model.fit(train_x, pred>0.5                     ,batch_size=16                     ,epochs=10                     ,verbose=2                     ,validation_split=0.2                     ,use_multiprocessing=True                    ) <\/code><\/pre>\n<pre><code class=\"python\">Epoch 1\/10 103\/103 - 3s - loss: 0.0045 - dice_coef: 0.9899 - val_loss: 0.0021 - val_dice_coef: 0.9944 Epoch 2\/10 103\/103 - 3s - loss: 0.0030 - dice_coef: 0.9933 - val_loss: 0.0017 - val_dice_coef: 0.9953 Epoch 3\/10 103\/103 - 3s - loss: 0.0027 - dice_coef: 0.9941 - val_loss: 0.0016 - val_dice_coef: 0.9958 Epoch 4\/10 103\/103 - 3s - loss: 0.0024 - dice_coef: 0.9946 - val_loss: 0.0016 - val_dice_coef: 0.9960 Epoch 5\/10 103\/103 - 3s - loss: 0.0022 - dice_coef: 0.9951 - val_loss: 0.0014 - val_dice_coef: 0.9963 Epoch 6\/10 103\/103 - 3s - loss: 0.0021 - dice_coef: 0.9953 - val_loss: 0.0014 - val_dice_coef: 0.9963 Epoch 7\/10 103\/103 - 3s - loss: 0.0021 - dice_coef: 0.9955 - val_loss: 0.0014 - val_dice_coef: 0.9965 Epoch 8\/10 103\/103 - 3s - loss: 0.0020 - dice_coef: 0.9957 - val_loss: 0.0014 - val_dice_coef: 0.9965 Epoch 9\/10 103\/103 - 3s - loss: 0.0019 - dice_coef: 0.9958 - val_loss: 0.0013 - val_dice_coef: 0.9967 Epoch 10\/10 103\/103 - 3s - loss: 0.0019 - dice_coef: 0.9959 - val_loss: 0.0014 - val_dice_coef: 0.9965<\/code><\/pre>\n<p>\u0418 \u0442\u0443\u0442 \u043f\u043e\u043b\u0443\u0447\u0430\u0435\u043c \u043f\u0440\u043e\u0441\u0442\u043e \u0444\u0430\u043d\u0442\u0430\u0441\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0439 \u0440\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442 &#8212; \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u044c \u0431\u043e\u043b\u044c\u0448\u0435 0.99! <\/p>\n<p>\u0410 \u043a\u043e\u043d\u043a\u0440\u0435\u0442\u043d\u043e \u0434\u043e\u0441\u0442\u0438\u0433\u0430\u0435\u043c dice_coeff 0.9919946, \u0438\u043c\u0435\u044f \u043d\u0430 \u0440\u0443\u043a\u0430\u0445 \u0434\u0440\u044f\u043d\u043d\u0443\u044e \u0440\u0430\u0437\u043c\u0435\u0442\u043a\u0443 \u0432 \u043d\u0430\u0447\u0430\u043b\u0435.<\/p>\n<pre><code>pred_1 = model.predict(train_x) dice_coef(train_r.astype('float32'),<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[],"tags":[],"class_list":["post-339405","post","type-post","status-publish","format-standard","hentry"],"_links":{"self":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts\/339405","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=339405"}],"version-history":[{"count":0,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=\/wp\/v2\/posts\/339405\/revisions"}],"wp:attachment":[{"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=339405"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=339405"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/savepearlharbor.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=339405"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}